基于DHMM和BP神经网络的齿轮箱故障诊断研究

摘要

为提高齿轮箱故障的智能诊断精度,从信息融合的角度,提出了一种基于DHMM和BP神经网络的混合智能故障诊断方法.根据不同工况下齿轮箱的振动信号时频特征,利用训练样本建立各类工况下的DHMM模型,然后求得测试样本在各DHMM模型下的似然概率对数,将似然概率对数作为新的特征添加到原来时频特征中,把新的特征集作为BP神经网络的输入,实现各工况的诊断.实验结果证明,相比于单独使用DHMM方法、BP神经网络以及两种方法的简单级联,该方法较大的提高了齿轮箱故障的诊断精度.将DHMM方法引入到齿轮箱的故障诊断中,结合了BP神经网络的自适应能力强和DHMM时序建模能力强的优点,具有一定的应用价值.%In order to improve the accuracy of intelligent diagnosis on gearbox faults, from the perspective of information fusion,proposes a hybrid intelligent fault diagnosis method based on DHMM and BP neural network. According to the time-frequency characteristics abstracted from the vibration signal of the gearbox under different working conditions, the DHMM models under various working conditions were established by using the training samples. Then,calculated the probability logarithm of the test samples in each DHMM model,added the probability logarithm as new features to the original time-frequency characteristics, took the new feature set as BP neural network input to realize the diagnosis of different working conditions. The experimental results show that, compared with DHMM method, BP neural network alone and their simple cascading,this method can improve the diagnosis accuracy of gearbox fault greatly. Introduces DHMM into fault diagnosis of gearbox,combines the self-adaptive ability of BP neural network and the timing modeling ability of DHMM method,and has certain practical value.